mordant 0.8.10

Mordant - A fast CommonMark + GFM Markdown parser for Python

Mordant

Version: 0.8.10
Rust: rushdown v0.18.0 (CommonMark 0.31.2 + GFM)
Python: 3.9+
Bindings: PyO3 0.29

A fast CommonMark + GFM Markdown parser and renderer for Python, powered by the rushdown Rust library.

• Architecture — Full architecture documentation
• Quick Reference — Python bindings quick reference

What's New in 0.8.10

• Server-side Mermaid rendering — Mermaid diagrams now render as inline SVG via the mermaid-rs-renderer crate (~3ms server-side vs ~2s client-side). No browser/CDN dependency. Three render modes: server (default, inline SVG), client (legacy, Mermaid.js ESM), hybrid (try server, fallback to client)
• Render mode API — PyDiagramHtmlRendererOptions(render_mode="server"|"client"|"hybrid", mermaid_url=...)
• Customizable Mermaid themes — Mermaid color schemes derived from code-highlighting (syntect) themes. PyDiagramHtmlRendererOptions(theme="Dracula") themes server-side SVG (via mermaid-rs-renderer) and client-side rendering (via mermaid.initialize + themeVariables). A single theme= kwarg on markdown_to_html themes both code and diagrams; native mermaid themes (modern/dark/forest/neutral) are also supported.

What's New in 0.8.7

• Chunker GFM + Diagram parity — MarkdownChunker now uses the same parser extensions as parse() and markdown_to_html(): GFM tables (TableAstTransformer, TableParagraphTransformer) and Mermaid diagrams (DiagramAstTransformer) are correctly classified as BlockType::Table and BlockType::Diagram respectively
• Chunker Diagram block type — BlockType::Diagram added to the chunker's type system; mermaid code blocks yield as "Diagram" instead of "CodeBlock" or being silently dropped
• Diagram source position fix — DiagramAstTransformer now copies the original code block's pos() to the new Diagram node so the chunker can slice raw source correctly

What's New in 0.8.6

• Lint engine — 25 lint rules (MD001, MD003, MD009, MD010, MD012, MD013, MD018–MD022, MD024, MD025, MD026, MD031, MD032, MD034, MD040, MD042, MD045–MD048, MD049, MD050) with diagnostics, fix engine, and configuration
• Batch API — lint_many() and fix_many() for parallel file processing via rayon, with GIL release for the entire batch
• CLI — python -m mordant with --fix, --dry-run, --format (human/json/github), --config, --enable, --disable, --default-language, glob/directory recursion
• Phase 8 accuracy polish — emoji text in heading comparison (MD024), frontmatter title: support (MD025), fragment anchor validation for links (MD042)
• Document chunking — MarkdownChunker lazy AST-based chunk iterator yielding bare chunks (no heading prefix), with get_chunks(), get_all_chunks(), get_chunks_with_context(), get_bare_chunks(), ExtractedChunk (with block_type/start_offset/end_offset), get_delimiter(), compute_overlap_payloads()
• Inline suppression — <!-- markdownlint-disable MD001 --> comments supported
• VSCode JSON theme support — Custom themes from .json files via add_custom_theme() and user directory ~/.mordant/themes/
• 1297 tests passing (up from 1161)

Features

• Blazing fast. One of the fastest Markdown parsers for Python — up to 55x faster than python-markdown on large documents.
• Full AST access. Parse markdown to a Document with complete tree traversal — navigate parent, children, siblings, access all node kinds.
• CommonMark + GFM. Fully compliant with CommonMark 0.31.2 and GitHub Flavored Markdown (tables, task lists, strikethrough; autolink disabled by default, enable with GfmOptions.all()).
• YAML frontmatter. Extract metadata from YAML frontmatter with full type preservation (null, bool, int, float, str, list, dict).
• Multi-threaded. Parse and render release the GIL — scale ~4.0x linearly with thread count.
• Emoji support. :joy: :heart: :smile: — shortcode-style emoji rendering with blacklist and custom templates.
• Math support. LaTeX math via KaTeX — fenced math/latex blocks, inline $...$/$$...$$ math, standalone render_math() function.
• Mermaid diagrams. graph LR, sequenceDiagram — render Mermaid diagrams from code blocks, server-side as inline SVG by default. Customizable color schemes derived from code-highlighting themes (a single theme= kwarg themes both code and diagrams).
• Footnotes. PHP Markdown Extra style footnotes ([^1], [^hello]) with <sup> references, <div class="footnotes"> endnotes, and backlinks.
• Document chunking. MarkdownChunker — lazy, low-copy AST-based chunk iterator with heading-context propagation, from_file() and from_file_mmap() constructors.
• Extensible. Custom node types, parsers, transformers, and renderers via Rust extensions.

Install

pip install mordant

Or from source:

cd mordant-py
cargo build --release
pip install -e .

Quick Start

import mordant

# Parse + render in one call
html = mordant.markdown_to_html("# Hello\n\n**World**")
# '<h1>Hello</h1>\n<p><strong>World</strong></p>\n'

# GFM support (tables, strikethrough, task lists enabled by default)
html = mordant.markdown_to_html("~~deleted~~")
# '<p><del>deleted</del></p>\n'

# Autolink (disabled by default; enable with GfmOptions.all())
html = mordant.markdown_to_html(
    "https://example.com",
    gfm_opts=mordant.GfmOptions.all()
)
# '<p><a href="https://example.com">https://example.com</a></p>\n'

# Full AST access
doc = mordant.parse("# Hello\n\n**World**")
print(doc.kind)        # "Document"
print(doc.children)    # [Heading, Paragraph]
print(doc.text)        # "HelloWorld"

# Emoji support
html = mordant.markdown_to_html("I'm :joy: and :heart:")
# '<p>I'm 😀 and ❤️</p>\n'

# Emoji blacklist
opts = mordant.PyEmojiParserOptions(blacklist="joy")
html = mordant.markdown_to_html(":joy: :heart:", emoji_parse_opts=opts)
# ':joy:' passes through; :heart: renders as ❤️

# Math support
html = mordant.markdown_to_html("""```math
\\int_0^\\infty e^{-x^2} dx = \\frac{\\sqrt{\\pi}}{2}
```""")
# '<span class="katex katex-display">...</span>'

# Standalone math rendering
result = mordant.render_math(r"\alpha + \beta", display=True, output="both")

# Footnotes (always enabled)
html = mordant.markdown_to_html("Text[^1]\n\n[^1]: The footnote.")
# '<p>Text<sup id="fnref:1"><a href="#fn:1" class="footnote-ref">1</a></sup></p>\n<div class="footnotes" role="doc-endnotes">\n<hr>\n<ol><li id="fn:1">The footnote.&#160;<a href="#fnref:1" class="footnote-backref" role="doc-backlink">&#x21a9;&#xfe0e;</a></li></ol></div>'

# Custom footnote options
opts = mordant.PyFootnoteHtmlRendererOptions(
    link_class="my-ref",
    backlink_class="my-back",
    backlink_html="↑ back",
)
html = mordant.markdown_to_html("Text[^1]", footnote_render_opts=opts)

# Mermaid diagrams
html = mordant.markdown_to_html("""```mermaid
graph LR
    A --- B
```""")
# '<pre class="mermaid">\ngraph LR\n    A --- B\n</pre>\n<script type="module">...'

# Mermaid with custom URL
opts = mordant.PyDiagramHtmlRendererOptions(mermaid_url="https://cdn.example.com/mermaid.mjs")
html = mordant.markdown_to_html("""```mermaid
graph TD
    A --> B
```""", diagram_render_opts=opts)

# Themed Mermaid diagram — color scheme derived from a code-highlighting theme
opts = mordant.PyDiagramHtmlRendererOptions(render_mode="server", theme="Dracula")
html = mordant.markdown_to_html("""```mermaid
graph TD
    A --> B
```""", diagram_render_opts=opts)
# Server-rendered SVG uses Dracula's palette (background #282a36, pink edges, ...)

# Single `theme=` kwarg themes BOTH code blocks and Mermaid diagrams
html = mordant.markdown_to_html(
    "# Title\n```mermaid\ngraph LR\n A---B\n```\n```python\nx=1\n```",
    theme="Dracula",
)
# Code block and diagram share Dracula's colors

# YAML frontmatter
md = """---
title: My Doc
author: Jane
tags: [rust, markdown]
---

Body
"""
doc = mordant.parse(md)
print(doc.metadata)
# {'title': 'My Doc', 'author': 'Jane', 'tags': ['rust', 'markdown']}

# Document chunking
chunker = mordant.MarkdownChunker("# Section\n\nPara one\n\n## Sub\n\nPara two")
for chunk in chunker:
    print(chunk)
# Para one
# Para two
# (bare chunks — no heading prefix)

# get_chunks() returns ExtractedChunk with metadata
for chunk in chunker.get_chunks():
    print(chunk.block_type, chunk.text, chunk.start_offset, chunk.end_offset)
# Paragraph Para one 9 17
# Paragraph Para two 27 35

# get_all_chunks() includes headings
for chunk in chunker.get_all_chunks():
    print(chunk.block_type, chunk.text)
# Heading # Section
# Paragraph Para one
# Heading ## Sub
# Paragraph Para two

# get_chunks_with_context() adds heading prefix
for chunk in chunker.get_chunks_with_context():
    print(chunk.text)
# # Section\n\nPara one
# ## Sub\n\nPara two

# compute_overlap_payloads() for embedding
payloads = chunker.compute_overlap_payloads(2)
# [{"chunk:0": "Para one"}, {"chunk:1": "one\n\nPara two"}]

Document Chunking

Split a document into bare chunks — each chunk is the raw block content with no heading prefix. OKF injects heading context at embed time for better embeddings. Headings update a current_header context; thematic breaks and other non-body nodes are skipped without resetting context.

import mordant

# Basic chunking — bare chunks (no heading prefix)
chunker = mordant.MarkdownChunker("# Section\n\nPara one\n\n## Sub\n\nPara two")
chunks = list(chunker)
assert len(chunks) == 2
assert chunks[0] == "Para one"          # bare, no heading prefix
assert chunks[1] == "Para two"          # bare, no heading prefix

# current_header still tracks the last heading seen
assert chunker.current_header == "## Sub"

# get_chunks() returns ExtractedChunk with metadata
for chunk in chunker.get_chunks():
    print(chunk.block_type, chunk.text, chunk.start_offset, chunk.end_offset)
# Paragraph Para one 9 17
# Paragraph Para two 27 35

# get_all_chunks() includes headings as separate chunks
for chunk in chunker.get_all_chunks():
    print(chunk.block_type, chunk.text)
# Heading # Section
# Paragraph Para one
# Heading ## Sub
# Paragraph Para two

# get_chunks_with_context() adds heading prefix for display
for chunk in chunker.get_chunks_with_context():
    print(chunk.text)
# # Section\n\nPara one
# ## Sub\n\nPara two

# get_delimiter() for document reconstruction
mordant.MarkdownChunker.get_delimiter("List", "List")           # "\n"
mordant.MarkdownChunker.get_delimiter("Blockquote", "Blockquote")  # "\n> "
mordant.MarkdownChunker.get_delimiter("Paragraph", "CodeBlock")  # "\n\n"

# compute_overlap_payloads() for embedding context continuity
chunker = mordant.MarkdownChunker("# Title\n\nFirst para second para third para.\n\n## Sub\n\nMore text here.")
payloads = chunker.compute_overlap_payloads(2)
# [{"chunk:0": "First para second para third para."},
#  {"chunk:1": "third  para.\n\nMore text here."}]

# from_file reads from disk
chunker = mordant.MarkdownChunker.from_file("/path/to/doc.md")
for chunk in chunker:
    print(chunk)  # bare chunks, no heading prefix

# from_file_mmap for zero-copy large files
chunker = mordant.MarkdownChunker.from_file_mmap("/path/to/large.md")

# Nested headings inside blockquotes never leak as context
chunker = mordant.MarkdownChunker("# Outer\n\n> # Nested\n\n> Quote text.")
chunks = list(chunker)
# current_header is "# Outer" (not "# Nested" which is nested)
assert chunker.current_header == "# Outer"

See QUICKREF.md for full API reference.

AST Traversal

doc = mordant.parse("# Title\n\n**Bold** and *italic*")

# Navigate tree
heading = doc.children[0]
print(heading.level)       # 1
print(heading.text)        # "Title"

# Walk all nodes
for node in doc.walk("depth"):
    print(f"{node.kind}: {node.text}")

# Find by kind
links = [n for n in doc.walk("depth") if n.kind == "Link"]

Options

# Parse options
parse_opts = mordant.ParseOptions(
    attributes=False,
    auto_heading_ids=False,
    escaped_space=False,
    meta_table=False,
)

# Render options
render_opts = mordant.RenderOptions(
    hard_wraps=False,
    xhtml=False,
    allows_unsafe=False,
    escaped_space=False,
)

# GFM options (default: tables + strikethrough + task lists; linkify disabled)
import mordant

gfm_opts = mordant.GfmOptions()
# Enable all features including linkify
gfm_opts = mordant.GfmOptions.all()
# Granular feature selection
gfm_opts = mordant.GfmOptions(features=[
    mordant.GfmFeature.Table,
    mordant.GfmFeature.Strikethrough,
])

html = mordant.markdown_to_html(
    "Hello\nWorld",
    gfm_opts=gfm_opts,
    parse_opts=parse_opts,
    render_opts=render_opts,
)

Multi-threaded Usage

from concurrent.futures import ThreadPoolExecutor
import mordant

# GIL is released during parse + render — safe for concurrent use
with ThreadPoolExecutor(max_workers=4) as pool:
    results = list(pool.map(mordant.markdown_to_html, markdown_docs))
# ~4.0x linear scaling vs single-threaded

Performance

Single-threaded (50 iterations)

Fixture	mordant	mistune	markdown-it-py	python-markdown
Small (400B)	0.039ms	0.430ms	0.475ms	2.301ms
Medium (5.4KB)	0.155ms	2.448ms	3.940ms	6.455ms
Large (26.7KB)	0.410ms	8.611ms	16.743ms	31.304ms
Data (202KB)	2.763ms	38.152ms	65.736ms	621.295ms

Multi-threaded (4 threads, medium fixture)

Library	1-thread	4-threads	Scaling
mordant	~1,000 docs/s	~4,000 docs/s	4.0x
python-markdown	~59 docs/s	~257 docs/s	4.35x
mistune	~133 docs/s	~542 docs/s	4.07x
markdown-it-py	~83 docs/s	~337 docs/s	4.06x

Node Kind Reference

Kind	Type	Example
Document	block	Root node
Paragraph	block	Hello world
Heading	block	# Title
ThematicBreak	block	---
CodeBlock	block	```python ... ```
Blockquote	block	> quoted
List	block	- item
ListItem	block	- [x] done
HtmlBlock	block	<div>...</div>
Text	inline	Plain text
CodeSpan	inline	`code`
Emphasis	inline	*italic*
Strong	inline	**bold**
Link	inline	[text](url)
Image	inline	
RawHtml	inline	<span>
LinkReferenceDefinition	block	[ref]: url
Table	block	`
TableHeader	block	Header row
TableBody	block	Body rows
TableRow	block	<tr>
TableCell	block	<td>
Strikethrough	inline	~~text~~
Diagram	block	```mermaid ... ```
FootnoteReference	inline	[^1], [^hello]
FootnoteDefinition	block	[^1]:, [^hello]:
Extension	any	Custom nodes

Thematic Break vs Frontmatter

The meta parser uses lookahead to distinguish --- (thematic break) from frontmatter:

# Thematic break
mordant.parse("---").metadata == {}

# Frontmatter
mordant.parse("---\ntitle: Test\n---").metadata["title"] == "Test"

# Five dashes is thematic break
mordant.parse("-----").metadata == {}

Error Handling

import mordant

try:
    doc = mordant.parse("---\ninvalid: yaml: [broken")
    doc.metadata  # Raises ValueError on access
except ValueError as e:
    print(e)  # YAML parsing error message

Architecture

Mordant wraps the rushdown Rust library (CommonMark 0.31.2 + GFM) via PyO3 bindings:

• Rust core: rushdown v0.18.0 — arena-allocated AST, priority-based parser dispatch, HTML renderer
• Python bindings: PyO3 0.29 — Document, Node, Walker classes with shared Rc<RefCell<Arena>> and Rc<str> source memory model (refcount bump on node creation instead of deep source copy)
• GIL release: Parse and render release the GIL via Python::detach() for multi-threaded parallelism
• Frontmatter: YAML parsing via yaml-peg with thematic break conflict resolution

rushdown-meta

YAML frontmatter support is provided by rushdown-meta, which has been directly incorporated into mordant. The original rushdown-meta crate is available in extensions/rushdown-meta-main/.

Key features of the integrated meta parser:

• Thematic break conflict resolution: --- alone is a thematic break; ---\n + YAML-like content is frontmatter
• Full YAML subset: null, bool, int, float, str, list, dict (via yaml-peg)
• AST table rendering: Optional meta_table option renders metadata as an HTML table in the AST
• Error handling: YAML parse errors are inserted as HTML comments in the AST; Python raises ValueError on doc.metadata access

See ARCHITECTURE.md §6 for full details.

rushdown-emoji

Emoji shortcode support (:joy:, :heart:, :smile:, etc.) is provided by rushdown-emoji, which has been directly incorporated into mordant. The original rushdown-emoji crate is available in extensions/rushdown-emoji-main/.

Key features of the integrated emoji extension:

• Shortcode parsing: :joy: → 😀, :heart: → ❤️, 1,500+ emojis from the emojis crate (v0.8.0)
• Blacklist support: PyEmojiParserOptions(blacklist="joy,heart") — blacklisted shortcodes pass through as literal text
• Custom HTML templates: PyEmojiHtmlRendererOptions(template='<img src="{shortcode}.png" />') — render emojis as <img> tags or any custom format
• Template placeholders: {emoji} (Unicode char), {shortcode} (e.g. "joy"), {name} (e.g. "grinning face with smiling eyes")
• Code span protection: Emojis inside `code` are not parsed — :joy: stays literal in code spans
• AST node access: Emoji nodes expose emoji, shortcode, and name properties via the Extension node kind
• Error handling: Unknown shortcodes pass through as-is (:invalid: → :invalid:)

See ARCHITECTURE.md §7.10 for full details.

rushdown-diagram

Diagram support is provided by rushdown-diagram, which has been directly incorporated into mordant. The original rushdown-diagram crate is available in extensions/rushdown-diagram-main/.

rushdown-diagram supports two diagram formats:

• MermaidJS — client-side rendering via the Mermaid.js ESM module
• PlantUML — server-side rendering (requires a plantuml command)

Mordant currently implements Mermaid support only. Key features:

• Code block detection: ```mermaid code blocks are automatically detected and converted to diagram nodes via an AST transformer
• Client-side rendering: Diagrams render as <pre class="mermaid"> with automatic Mermaid.js ESM script injection (single script tag for all diagrams)
• Custom Mermaid URL: PyDiagramHtmlRendererOptions(mermaid_url="https://cdn.example.com/mermaid.mjs") — use a custom Mermaid.js CDN or local file
• Customizable themes: PyDiagramHtmlRendererOptions(theme="<name>") derives Mermaid colors from a code-highlighting (syntect) theme — server-side SVG via render_with_options, client-side via mermaid.initialize + themeVariables. Built-in mermaid themes (modern/dark/forest/neutral) are used natively. A single theme= kwarg on markdown_to_html themes both code and diagrams; explicit per-param args override it.
• Parser options: PyDiagramParserOptions(mermaid_enabled=False) — disable diagram transformation to keep code blocks as regular fenced code blocks
• AST node access: Diagram nodes expose diagram_type ("mermaid") and diagram_value (source content) properties via the Diagram node kind
• Multiple diagrams: Multiple Mermaid blocks in one document all render correctly with a single script tag
• GFM compatible: Works alongside other GFM features (tables, task lists, strikethrough; autolink disabled by default, enable with GfmOptions.all())
• Frontmatter compatible: Works alongside YAML frontmatter

See ARCHITECTURE.md for full details.

rushdown-math

Math support is provided by the pure-Rust katex-rs crate, incorporated directly into mordant.

Key features:

• Fenced math blocks: ```math and ```latex code blocks render to KaTeX markup
• Inline math: $...$ for inline, $$...$$ for display mode
• Standalone render_math(): mordant.render_math(r"\alpha + \beta", display=True, output="both") — renders LaTeX independently of the Markdown AST
• Output formats: "both" (HTML+MathML, default), "html", or "mathml"
• Error handling: Invalid LaTeX produces an error span (<span class="katex-error">...) instead of crashing
• Caching: Rendered markup is memoized on (display, output, latex) for repeated formulas
• GIL released: Math rendering runs with the GIL released for multi-threaded parallelism

See ARCHITECTURE.md §7.12 for full details.

rushdown-footnote

Footnote support is provided by rushdown-footnote, which has been directly incorporated into mordant. Footnotes are always enabled — no parser options to disable them.

Syntax (PHP Markdown Extra):

Text with a footnote.[^1]
Text with a named footnote.[^hello]

[^1]: The footnote.

[^hello]: The named footnote.

Output:

<p>Text with a footnote.<sup id="fnref:1"><a href="#fn:1" class="footnote-ref">1</a></sup></p>
<div class="footnotes" role="doc-endnotes">
<hr>
<ol>
<li id="fn:1">The footnote.&#160;<a href="#fnref:1" class="footnote-backref" role="doc-backlink">&#x21a9;&#xfe0e;</a></li>
</ol>
</div>

Key features:

• Inline references: [^1], [^hello] — rendered as <sup><a href="#fn:N">N</a></sup>
• Block definitions: [^1]: followed by content — rendered in <div class="footnotes"> at end of document
• Named footnotes: [^hello] — label preserved in ID
• Multiple refs: Multiple [^1] to same [^1]: — each gets a superscript ref, definition rendered once
• Backlinks: Each definition has a backlink anchor (↩︎) to return to the reference
• Accessibility: role="doc-endnotes", role="doc-noteref", role="doc-backlink" ARIA attributes
• Custom options: PyFootnoteHtmlRendererOptions for custom CSS classes, backlink HTML, and ID prefixes
• AST node access: node.footnote_label, node.footnote_index, node.footnote_references properties
• No parser options: Footnotes are always enabled (matches math extension pattern)

See ARCHITECTURE.md §7.14 for full details.

Benchmarks

Run benchmarks:

cd mordant-py
python benchmarks.py              # All fixtures, 50 iterations
python benchmarks.py -f medium -n 100  # Specific fixture, custom count
python benchmarks.py -o results.json  # Save JSON

Tests

cd mordant-py
python -m pytest tests/ -v

1233 Python tests passing (Core, AST, GFM, Options, YAML Frontmatter, Emoji, Mermaid Diagrams, Math, Lint engine, CLI, batch API, Phase 8 accuracy, VSCode theme, Chunker, OKF chunker methods, Extracted Chunk, Mixed Features) + 64 Rust tests (Unit tests, AST, CommonMark spec, Extensions, GFM, Options, Doc-tests).

Theme Loading

Themes are loaded from multiple sources:

• Embedded themes — Bundled in mordant/themes/, loaded at import time
• User themes — Place .json or .tmTheme files in ~/.mordant/themes/ (or %APPDATA%/mordant/themes/ on Windows) for auto-loading
• Built-in themes — Loaded from syntect-assets (bat's updated themes)
• Custom themes — Use add_custom_theme(name, content) to register themes from JSON or XML content

Both VSCode JSON and Sublime .tmTheme formats are supported. VSCode JSON themes are automatically converted to the syntect format via the parse_vscode_theme_jsonc → vscode_theme_to_syntect pipeline, allowing you to use any VSCode theme file directly.

See QUICKREF.md for details.

License

MIT

Author

Mordant: Python bindings by opticsWolf Rushdown: Rust core by Yusuke Inuzuka